Correlation Analysis for initial propagation of COVID-19 in Argentina

Author: Denise Stefania Cammarota

Theoretical background and Motivation

During my master thesis on COVID-19 propagation, it was remarked that Argentinian provinces (administrative divisions similar to states) seemed to drive the epidemiological dynamics during the start of the pandemic in 2020. For example, the province corresponding to the Capital, as well as some northern provinces seemed to have peaks that preceded those of the rest of the country. This is interesting for many reasons like predicting the existence in peaks in other places after observing the dynamics of a province, or predicting which connections between provinces are more important during the propagation of COVID-19 (or any other disease). Therefore, we started to study whether it was possible to quantify this type of relationship between provinces. To do that, we chose to use lagged cross-correlations of time series of different provinces. This is a standard way of computing relationships between heterogeneous epidemic dynamics, specially with small datasets. As a result, we expect to get matrices of correlations and lags between provinces.

Even though I have done similar analyses in Python, I found that this project would be of great use to me in the light of a few things. Firstly, upon taking the Introduction to Scientific Programming/Computational Methods course, I realized that many of my problems when I did this in my thesis were due to the lack of a correctly organized workflow. Secondly, since I am currently writing a paper on this work, I think this process would greatly benefit from better organization of data and codes, as well as a second look at my analyses. Last, but not least, I am not a proficient R coder and I learn a lot by doing, so this seems like a great opportunity to learn more R.

Goals

My main goals for this project include:

• Process the raw data of new COVID-19 cases in Argentina using R.
• Separate the processed data into the corresponding time series of infected per province.
• Calculate the lagged cross-correlations for all pairs of provinces.
• Plot and analyse those lags and correlations to extract insights such as:
  • Which are the biggest correlations and lags? Does that make sense?
  • Which provinces seem to drive epidemic dynamics?
• If I have time before the deadline, I would like to repeat this analysis with deaths, since this is a much more clear indicator of epidemiological situation that does not depend on how many people decide to get tested or differences in symptoms required for testing between different administrative divisions in the country.

Methods

As it was previously stated in the first section, the relationship between the epidemiological dynamics of provinces is quantified by calculating lagged cross-correlations. In order to calculate these quantities in R, we use the function ccf that is present in base R, indicating that the maximum possible lag is around 365 days. That is, the lag between two provinces could have any value between 0 and 365 days.

Dataset

The dataset of confirmed cases is the official dataset provided by the Argentinian Ministry of Health to this day. The dataset is periodically updated in this link. However, the one I am using was downloaded from that same link on Saturday 13th of August of 2022, and can be found in this link of my personal Drive, since the official sites updates the dataset but leaves no record of previous ones. Even though I use my local copy of the dataset to perform all analysis, I haven't uploaded it to GitHub since its size is about 6 GB.

Requirements

Windows, Linux or MacOS with R installed, at least 8GB of RAM to explore and process the raw data. Apart from base R, the following libraries need to be installed in order to run all scripts:

dplyr
data.table
ggplot2
reshape2
matrixStats

These can be installed by running the following commands in an R terminal:

install.packages("dplyr")
install.packages("data.table")
install.packages("ggplot2")
install.packages("reshape2")
install.packages("matrixStats")

Project structure

This is the corresponding folder structure, updated as of the day of the last commit to the project. This structure is subject of modification until completion of the project.

project/
*    ├── data/
*    │   ├── raw
*    │   └── processed
*    ├── docs/
*    ├── fct/
*    ├── figs/
*    ├── outputs/
*    ├── R/
*    └── README.md

• The data folder contains two subfolders with raw and processed data. The raw data folder is not found on the repository, since it contains the total cases file, whose size is too big to be uploaded to GitHub. The processed subfolder contains processed cases and time series data.
• The docs folder contains markdown and html files. In this case, the resulting report can be found in this folder.
• The fct folder contains custom R functions in order to process and analyze the data.
• The figs folder contains figures obtained from data analysis.
• The outputs folder contains important outputs generated from our scripts. In this case, the time series of provinces with the corresponding information on provinces names and dates, along with the correlations and lags matrices can be found here.
• The R folder contains the R script to process and analyze the data.